metadata
license: apache-2.0
language:
- et
pipeline_tag: automatic-speech-recognition
This model is ONNX version of an Icefall Zipformer streaming ASR model for Estonian. It is identical in training details to https://huggingface.co/TalTechNLP/streaming-zipformer-large.et-en, but it's trained on transcript where number expressions are converted from words to digits.
Note that it tends to struggle with more complex numerical expressions (exceeding four digits).
Use sherpa-onnx to do ASR:
E.g., under Linux, using card 3 as input device:
sherpa-onnx-alsa --encoder=encoder.onnx --decoder=decoder.onnx --joiner=joiner.onnx --tokens=tokens.txt --decoding-method=modified_beam_search plughw:3,0